Message ID | 20210922133116.102-1-richard.gong@amd.com (mailing list archive) |
---|---|
State | Mainlined, archived |
Headers | show |
Series | [PATCHv1] ACPI: processor idle: Allow playing dead in C3 state | expand |
On Wed, Sep 22, 2021 at 3:31 PM Richard Gong <richard.gong@amd.com> wrote: > > When some cores are disabled on AMD platforms, the system will no longer > be able to enter suspend-to-idle s0ix. > > Update to allow playing dead in C3 state so that the CPUs can enter the > deepest state on AMD platforms. > > BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1708 > Suggested-by: Mario Limonciello <mario.limonciello@amd.com> > Signed-off-by: Richard Gong <richard.gong@amd.com> > --- > drivers/acpi/processor_idle.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c > index f37fba9e5ba0..61d5a72d218e 100644 > --- a/drivers/acpi/processor_idle.c > +++ b/drivers/acpi/processor_idle.c > @@ -789,7 +789,8 @@ static int acpi_processor_setup_cstates(struct acpi_processor *pr) > state->enter = acpi_idle_enter; > > state->flags = 0; > - if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { > + if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 > + || cx->type == ACPI_STATE_C3) { > state->enter_dead = acpi_idle_play_dead; > drv->safe_state_index = count; > } > -- Tentatively applied as 5.16 material, but have you done any research on why this restriction has been there in the first place?
Hi Rafael, On 9/24/2021 11:31 AM, Rafael J. Wysocki wrote: > On Wed, Sep 22, 2021 at 3:31 PM Richard Gong <richard.gong@amd.com> wrote: >> When some cores are disabled on AMD platforms, the system will no longer >> be able to enter suspend-to-idle s0ix. >> >> Update to allow playing dead in C3 state so that the CPUs can enter the >> deepest state on AMD platforms. >> >> BugLink: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.freedesktop.org%2Fdrm%2Famd%2F-%2Fissues%2F1708&data=04%7C01%7Crichard.gong%40amd.com%7Ca5c0db9ce02b4cd0864f08d97f78d33f%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637680980065602427%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=dIO5PDjpncLFIt0aGaTrKJHYIkJm8A4YByGn1%2FsHrtM%3D&reserved=0 >> Suggested-by: Mario Limonciello <mario.limonciello@amd.com> >> Signed-off-by: Richard Gong <richard.gong@amd.com> >> --- >> drivers/acpi/processor_idle.c | 3 ++- >> 1 file changed, 2 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c >> index f37fba9e5ba0..61d5a72d218e 100644 >> --- a/drivers/acpi/processor_idle.c >> +++ b/drivers/acpi/processor_idle.c >> @@ -789,7 +789,8 @@ static int acpi_processor_setup_cstates(struct acpi_processor *pr) >> state->enter = acpi_idle_enter; >> >> state->flags = 0; >> - if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { >> + if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 >> + || cx->type == ACPI_STATE_C3) { >> state->enter_dead = acpi_idle_play_dead; >> drv->safe_state_index = count; >> } >> -- > Tentatively applied as 5.16 material, but have you done any research > on why this restriction has been there in the first place? Yes, we need this change to align with the updated firmware on AMD platforms. Regards, Richard
On Tue, Sep 28, 2021 at 3:00 PM Gong, Richard <richard.gong@amd.com> wrote: > > Hi Rafael, > > On 9/24/2021 11:31 AM, Rafael J. Wysocki wrote: > > On Wed, Sep 22, 2021 at 3:31 PM Richard Gong <richard.gong@amd.com> wrote: > >> When some cores are disabled on AMD platforms, the system will no longer > >> be able to enter suspend-to-idle s0ix. > >> > >> Update to allow playing dead in C3 state so that the CPUs can enter the > >> deepest state on AMD platforms. > >> > >> BugLink: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.freedesktop.org%2Fdrm%2Famd%2F-%2Fissues%2F1708&data=04%7C01%7Crichard.gong%40amd.com%7Ca5c0db9ce02b4cd0864f08d97f78d33f%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637680980065602427%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=dIO5PDjpncLFIt0aGaTrKJHYIkJm8A4YByGn1%2FsHrtM%3D&reserved=0 > >> Suggested-by: Mario Limonciello <mario.limonciello@amd.com> > >> Signed-off-by: Richard Gong <richard.gong@amd.com> > >> --- > >> drivers/acpi/processor_idle.c | 3 ++- > >> 1 file changed, 2 insertions(+), 1 deletion(-) > >> > >> diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c > >> index f37fba9e5ba0..61d5a72d218e 100644 > >> --- a/drivers/acpi/processor_idle.c > >> +++ b/drivers/acpi/processor_idle.c > >> @@ -789,7 +789,8 @@ static int acpi_processor_setup_cstates(struct acpi_processor *pr) > >> state->enter = acpi_idle_enter; > >> > >> state->flags = 0; > >> - if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { > >> + if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 > >> + || cx->type == ACPI_STATE_C3) { > >> state->enter_dead = acpi_idle_play_dead; > >> drv->safe_state_index = count; > >> } > >> -- > > Tentatively applied as 5.16 material, but have you done any research > > on why this restriction has been there in the first place? > > Yes, we need this change to align with the updated firmware on AMD > platforms. Well, it wouldn't be necessary to change the code otherwise. Still, without the patch it is not allowed to play dead in deep idle states (C3 type) and do you know why? Or IOW how likely is this change to break anything on legacy platforms?
On 9/28/2021 7:18 PM, Rafael J. Wysocki wrote: > On Tue, Sep 28, 2021 at 3:00 PM Gong, Richard <richard.gong@amd.com> wrote: >> >> Hi Rafael, >> >> On 9/24/2021 11:31 AM, Rafael J. Wysocki wrote: >>> On Wed, Sep 22, 2021 at 3:31 PM Richard Gong <richard.gong@amd.com> wrote: >>>> When some cores are disabled on AMD platforms, the system will no longer >>>> be able to enter suspend-to-idle s0ix. >>>> >>>> Update to allow playing dead in C3 state so that the CPUs can enter the >>>> deepest state on AMD platforms. >>>> >>>> BugLink: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.freedesktop.org%2Fdrm%2Famd%2F-%2Fissues%2F1708&data=04%7C01%7Clijo.lazar%40amd.com%7C44c85e9ae1dd4bec087e08d98286aa72%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637684337168161919%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=Qoh%2BwD%2B1JzPYBc0rrGfNCdQWY1pKo8WBhLPmsmQ6G%2Bo%3D&reserved=0 >>>> Suggested-by: Mario Limonciello <mario.limonciello@amd.com> >>>> Signed-off-by: Richard Gong <richard.gong@amd.com> >>>> --- >>>> drivers/acpi/processor_idle.c | 3 ++- >>>> 1 file changed, 2 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c >>>> index f37fba9e5ba0..61d5a72d218e 100644 >>>> --- a/drivers/acpi/processor_idle.c >>>> +++ b/drivers/acpi/processor_idle.c >>>> @@ -789,7 +789,8 @@ static int acpi_processor_setup_cstates(struct acpi_processor *pr) >>>> state->enter = acpi_idle_enter; >>>> >>>> state->flags = 0; >>>> - if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { >>>> + if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 >>>> + || cx->type == ACPI_STATE_C3) { >>>> state->enter_dead = acpi_idle_play_dead; >>>> drv->safe_state_index = count; >>>> } >>>> -- >>> Tentatively applied as 5.16 material, but have you done any research >>> on why this restriction has been there in the first place? >> >> Yes, we need this change to align with the updated firmware on AMD >> platforms. > > Well, it wouldn't be necessary to change the code otherwise. > > Still, without the patch it is not allowed to play dead in deep idle > states (C3 type) and do you know why? > Found the original patch thread, unfortunately there is no discussion on why C3 was skipped or not allowed - https://www.spinics.net/lists/linux-pm/msg27003.html Maybe, Intel CPU offlining worked through MWAITs and AMD bothered only till C2. Thanks, Lijo > Or IOW how likely is this change to break anything on legacy platforms? >
On 9/28/2021 8:38 PM, Lazar, Lijo wrote: > > > On 9/28/2021 7:18 PM, Rafael J. Wysocki wrote: >> On Tue, Sep 28, 2021 at 3:00 PM Gong, Richard <richard.gong@amd.com> >> wrote: >>> >>> Hi Rafael, >>> >>> On 9/24/2021 11:31 AM, Rafael J. Wysocki wrote: >>>> On Wed, Sep 22, 2021 at 3:31 PM Richard Gong <richard.gong@amd.com> >>>> wrote: >>>>> When some cores are disabled on AMD platforms, the system will no >>>>> longer >>>>> be able to enter suspend-to-idle s0ix. >>>>> >>>>> Update to allow playing dead in C3 state so that the CPUs can enter >>>>> the >>>>> deepest state on AMD platforms. >>>>> >>>>> BugLink: >>>>> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.freedesktop.org%2Fdrm%2Famd%2F-%2Fissues%2F1708&data=04%7C01%7Clijo.lazar%40amd.com%7C44c85e9ae1dd4bec087e08d98286aa72%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637684337168161919%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=Qoh%2BwD%2B1JzPYBc0rrGfNCdQWY1pKo8WBhLPmsmQ6G%2Bo%3D&reserved=0 >>>>> >>>>> Suggested-by: Mario Limonciello <mario.limonciello@amd.com> >>>>> Signed-off-by: Richard Gong <richard.gong@amd.com> >>>>> --- >>>>> drivers/acpi/processor_idle.c | 3 ++- >>>>> 1 file changed, 2 insertions(+), 1 deletion(-) >>>>> >>>>> diff --git a/drivers/acpi/processor_idle.c >>>>> b/drivers/acpi/processor_idle.c >>>>> index f37fba9e5ba0..61d5a72d218e 100644 >>>>> --- a/drivers/acpi/processor_idle.c >>>>> +++ b/drivers/acpi/processor_idle.c >>>>> @@ -789,7 +789,8 @@ static int acpi_processor_setup_cstates(struct >>>>> acpi_processor *pr) >>>>> state->enter = acpi_idle_enter; >>>>> >>>>> state->flags = 0; >>>>> - if (cx->type == ACPI_STATE_C1 || cx->type == >>>>> ACPI_STATE_C2) { >>>>> + if (cx->type == ACPI_STATE_C1 || cx->type == >>>>> ACPI_STATE_C2 >>>>> + || cx->type == ACPI_STATE_C3) { >>>>> state->enter_dead = acpi_idle_play_dead; >>>>> drv->safe_state_index = count; >>>>> } >>>>> -- >>>> Tentatively applied as 5.16 material, but have you done any research >>>> on why this restriction has been there in the first place? >>> >>> Yes, we need this change to align with the updated firmware on AMD >>> platforms. >> >> Well, it wouldn't be necessary to change the code otherwise. >> >> Still, without the patch it is not allowed to play dead in deep idle >> states (C3 type) and do you know why? >> > > Found the original patch thread, unfortunately there is no discussion on > why C3 was skipped or not allowed - > https://www.spinics.net/lists/linux-pm/msg27003.html > > Maybe, Intel CPU offlining worked through MWAITs and AMD bothered only > till C2. > Hi Boris/Len, Apologies to barge in. This is regarding the patch you added years back - https://www.spinics.net/lists/linux-pm/msg27003.html Do you recall why C3 was not considered for offline/play dead scenarios in the original patch? For newer AMD processors, power management controller expects all cores to request C3 as a precondition for s2idle. When cores are offlined, they request only C2 and that prevents s2idle in such cases. This new patch series is to consider C3 also for offline purpose. https://lore.kernel.org/linux-acpi/7b8a7c2a-027f-16c9-0258-d91917c39bac@amd.com/T/#ma818870a8272a3755de896dd79d478cf944d1555 Please take a look and also at the questions from Rafael. Thanks, Lijo >> Or IOW how likely is this change to break anything on legacy platforms? >>
On 9/29/21 1:37 AM, Lazar, Lijo wrote: > > > > Apologies to barge in. This is regarding the patch you added years back - https://www.spinics.net/lists/linux-pm/msg27003.html > > Do you recall why C3 was not considered for offline/play dead scenarios in the original patch? Not really. Perhaps processors from that time (at least AMD ones) did not support C3? -boris
On Wed, Sep 29, 2021 at 4:09 PM Boris Ostrovsky <boris.ostrovsky@oracle.com> wrote: > > > On 9/29/21 1:37 AM, Lazar, Lijo wrote: > > > > > > > > Apologies to barge in. This is regarding the patch you added years back - https://www.spinics.net/lists/linux-pm/msg27003.html > > > > Do you recall why C3 was not considered for offline/play dead scenarios in the original patch? > > > Not really. Perhaps processors from that time (at least AMD ones) did not support C3? OK, thanks! I guess the only way to find out is to try it and see then.
diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c index f37fba9e5ba0..61d5a72d218e 100644 --- a/drivers/acpi/processor_idle.c +++ b/drivers/acpi/processor_idle.c @@ -789,7 +789,8 @@ static int acpi_processor_setup_cstates(struct acpi_processor *pr) state->enter = acpi_idle_enter; state->flags = 0; - if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { + if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 + || cx->type == ACPI_STATE_C3) { state->enter_dead = acpi_idle_play_dead; drv->safe_state_index = count; }
When some cores are disabled on AMD platforms, the system will no longer be able to enter suspend-to-idle s0ix. Update to allow playing dead in C3 state so that the CPUs can enter the deepest state on AMD platforms. BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1708 Suggested-by: Mario Limonciello <mario.limonciello@amd.com> Signed-off-by: Richard Gong <richard.gong@amd.com> --- drivers/acpi/processor_idle.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)